Electrical engineering is on the cusp of a breakthrough—one that will allow engineers to create circuits that drastically increase the speed of processing, use far less power than modern computers, and even mimic the kind of computations carried out by the human brain. This shift comes in the form of the “memristor,” a long-theorized but only recently constructed electrical component that stores information about its past activity and uses this information to influence its behavior. First theorized nearly 40 years ago, and finally built by Hewlett-Packard Labs in 2008, it promises to redefine the abilities and applications of computers of the future.
While memristors have only recently been constructed, they have existed in theoretical electronics for many years. Leon Chua, a longstanding member of the Department of Electrical Engineering and Computer Sciences at UC Berkeley, laid out the original theory in 1971. In his paper, Chua addressed a hole that existed in our knowledge of electrical engineering.
The world of electronics is largely built around devices that carry out interactions between the basic variables in any circuit: charge, resistance, voltage, and flux. For example, a capacitor creates a voltage by maintaining an imbalance of electrons (or charge) on either side of a gap. At the time of the theory’s publication, there was a clear explanation for how a real-world device could connect each combination of these elements but one: charge and flux.
Chua theorized a new circuit element to carry out the missing interaction. This element would behave very similarly to a resistor, but with one key difference: the amount that it impeded the flow of electricity would depend on the current that had already passed through. In essence, this electrical element would have a memory, combining information about the past with its input in the present. For this reason, Chua dubbed this new element the “memristor.”
Though it made a splash in theoretical electronics, it would be nearly 40 years until the memristor would be realized in the laboratory. Up to that point, Chua’s depiction of the properties of memristors had been likened to the elusive “Higgs boson” of theoretical physics: a particle that exists in theory but has not yet been observed. Then, in 2008, HP Labs announced that they had created a nano-scale circuit that showed exactly the same properties that Chua had theorized. Memristors were real.
Although memristors have yet to be successfully integrated into standard electronics, the ability to engineer circuits with memristors is improving rapidly, and hybrid memristor/traditional computers are expected to make their first appearance in consumer technology in the next few years. Your next computer could have memristors that allow for faster booting and processing. These early successes bode well for a paradigm shift in the future of electronics. While most modern computers perform calculations using dynamic random access memory (DRAM) that must be wiped clean every time a computer loses power, a new memristorequipped computer could “remember” the state from when it was last turned off and boot up nearly instantaneously.
Memristors could also decrease computers’ power consumption, which has increased exponentially as demands on processors continue to rise. Currently, this power consumption poses a significant challenge to increasing the complexity and power of processing chips. Memristors, however, consume relatively little power because storing memory within small units, rather than in a separate system, allows designers to use fewer and shorter wires, and thus less power. “Memristor systems bring data close to computation, much as biological systems do,” explains Massimiliano Versace, a researcher at Boston University who is using memristors to study, and possibly create, models that are inspired by human cognition.
The potential to create highly interconnected systems that are eerily similar to the way our own brains are structured is one of the most exciting potential applications for memristors. For many years, scientists have tried to model human cognition, but have often fallen short due to the limitations of our current hardware. Such systems are built with specific locations for computations (central processing units, or CPUs), short-term memory (dynamic random-access memory, or DRAM), and long-term memory (the hard drive). It is an inherently inefficient process, requiring lots of cross-talk and bottlenecks between these discrete areas.
Unfortunately for cognitive scientists (but fortunately for the rest of us), brains don’t work this way at all. And so our efforts to simulate brains have hit this fundamental roadblock: it is exceedingly difficult to create machines that act like brains without being built like them. Although artificial models of neural systems have grown remarkably in complexity and scope, artificial intelligence is still a far cry from resembling actual human brain function.
By allowing memory to be embedded directly within artificial networks, memristors are bringing cognitive science one critical step closer to mimicking the way that biological neural networks compute and store information. The CPU, DRAM, and hard drive of current computer systems can now be replaced with a constantly changing, interconnected system of simple memristor units, blurring the line between “computation” and “memory” and allowing engineers to produce more effective artificial intelligence models that truly mimic brain function.
At the forefront of this new approach to artificial cognition is the very man who first brought memristors into the public lexicon, Leon Chua. “Synapses are in fact memristors, axons are made of memristors, and thus, brains are made of memristors,” Chua suggests. At the foundation of his vision of a memristor brain is associative memory—in a nutshell, the ability to recognize an entire picture or idea when presented with only a small fraction of it. It is a process theorized to be a fundamental function of biological brains, and it requires retaining information about the past. In artificial circuits, a memristor does this well, essentially “storing” information about its previous activity, even when the circuit isn’t active, supporting Chua’s analogy.
Chua isn’t the only one interested in integrating memristors into brain-inspired circuits. Researchers at HP Labs and Boston University are already working toward a brain-like microprocessor based on memristor technology. “It will perceive its surroundings, decide which information is useful, integrate that information into the emerging structure of its reality, and in some applications, formulate plans that will ensure its survival,” write BU researchers Versace and Ben Chandler.
It has been only three years—an extremely short time in the world of technology—since the first memristors were created. What the future brings is anyone’s guess; memristors may eventually be used in ways that are unimaginable today. However, if Chua’s vision is realized, then they will certainly play a role in creating computers that are more powerful, more efficient, and maybe even more “human” than we can yet imagine.